AI专家激辩：AI失业、隐私终结与AI战争开端——Mo、Salim与Dave 176

In my mind, jobs will be lost. When they are lost, they're going to be lost massively.

Far more people are in denial or doing nothing than are overreacting. I do think governments are woefully underprepared.

Now, it comes down to this: are we going to design a world that is good for people or not?

We all know that AI will go out of control within the next 5 to 10 years. And yet we're building autonomous weapons after autonomous weapons, knowing for a fact that every other opponent anywhere in the globe is building them, too.

Trump taps Palantir to compile data on Americans. This is not a tech problem. This is an accountability problem.

What can I build that is making people happier and more productive, feeling valuable, and having a sense of purpose? If we focus on that, we actually can avoid the dystopian outcome.

We have the ability to create an intentional future. This future is not happening to us. We have the ability to guide where it goes.

I'm joined by three Moonshot mates today. Dave Blundin, the head of Link XP. Dave, good morning. Looks like you're at home today.

At home, yeah. Princeton graduation on Tuesday, MIT graduation on Thursday, and then off to Stanford tonight.

All right, fantastic. I look forward to seeing you, hopefully, this week. Salim Ismail, the CEO of ExO, OpenExO, and Sem[?]. Where do I find you today?

At home, just outside New York City, and looking forward to this episode.

Yeah, me too. And good morning—or good evening—Mo Gawdat, the one and only. Dubai? Is that where you are?

Dubai today, yes. Happy to be inside because it is boiling outside.

I'm in Santa Monica, just back from a few days in Hong Kong. It's crazy. You literally have no idea where your friends are these days. They're literally around the world. We're just tied together by this digital network of Zoom and a multitude of other technologies.

Anyway, it's been a crazy week in AI and in a whole slew of different technologies, and I'm excited to get into it. Before we start, anything new, Dave or Salim, you want to add?

First of all, thanks for getting up at 5:00 a.m. to do the podcast. It's hard to tie together Dubai and LA, but it is much appreciated. I'm in the middle zone here, so it's very easy for me, but thank you.

You're welcome. You guys are worth getting up for.

As always, every week is going at a pace that would have been unbelievable. I'm just trying to remember back 10 or 20 years ago—the number of breakthroughs or announcements that were occurring on a regular basis—and I can't find any analogy. I remember in the dot-com world there were all these crazy new dot-com companies being announced every week, but here it's not just crazy companies. It's fundamental capabilities that are coming online.

Go ahead.

We're also going to see later in the podcast a predicted $1 trillion a year of capital expenditure going forward. I checked, and that's the equivalent investment that we made mobilizing for World War II, inflation-adjusted. If it feels crazy, it should, because it's historic in scale.

All right, let's jump into a subject on a lot of people's minds. We've heard a lot of news about this: AI and job loss. We'll begin with a short segment of Dario Amodei, the CEO of Anthropic, talking about job loss. Let's take a listen and then discuss it.

I really worry, particularly at the entry level, that the AI models are very much at the center of what an entry-level human worker would do. I'm a little bit more worried about the labor impact simply because it's happening so fast that, yes, people will adapt, but they may not adapt fast enough. There may be an adjustment period in terms of inequality. I'm worried about this.

There's an inherent social contract in democracy where, ultimately, the ordinary person has a certain amount of leverage because they're contributing to the economy. If that leverage goes away, then it's hard to make democracy work. It's harder to make democracies work, and it's harder to prevent concentration of power.

We need to make sure that the ordinary person maintains economic leverage and has a way to make a living, or our society—our social contract—won't work. That's why you've previously described a future where cancer is cured, the economy grows at 10% a year, the budget is balanced, and 20% of people don't have jobs.

The quote you just flashed is maybe too optimistic, maybe too sanguine, about the ability for people to adapt. People have adapted to past technological changes, but I'll say again: everyone I've talked to has said, “This technological change looks different. It looks faster. It looks harder to adapt to. It's broader. The pace of progress keeps catching people off guard.”

I think the benefits are massive, and we need to find a way to achieve the benefits and mitigate or prevent the harms. The second thing I would say is, look, there are, as you mentioned, 6 or 7 companies in the US building this technology. If we stopped doing it tomorrow, the rest would continue. If all of us somehow stopped doing it tomorrow, then China would just beat us. I don't think China winning in this technology helps anyone or makes the situation any better.

All right, a lot there. We've heard this at an increasing pace and intensity. Dave, thoughts on Dario's commentary here?

Yeah, Dario looks really worried there, doesn't he? He's got the wrinkled forehead, and then Anderson Cooper looks even more worried. Rightfully so, I think the short-term job displacement is imminent.

When I talk to people at random—people in power—far more people are in denial or doing nothing than are overreacting. It's actually good for Dario to be saying these things to at least wake up the masses to the immense amount of change that's imminent. I do think there's a lot of short-term job loss coming, but far, far, far more opportunity being created.

It's kind of a foot race between creators and entrepreneurs reinventing what people do versus automators coming in and automating away white-collar jobs, and then, ultimately, robotics and blue-collar jobs.

Mo, you've been speaking about this for a while. Is Dario overplaying this, or is he spot-on?

No, he's underplaying it, for sure. My prediction is 10%, 20%, 30%, 40% unemployment in some sectors.

In what time frame?

Within the next 2 to 3 years. I think there are always 3 questions to answer. One is, does anyone on this call believe that the technology is not going to catch up for some of those jobs, like a graphic designer or a video editor, for example? Those sectors are gone.

Today, with Veo 3 giving you a minute of video that's better than Avatar for $17, you can create the movie Avatar for around $1,500 if you make mistakes along the way. I don't know how we can save those jobs, to be quite honest.

If that's the case, then the next question becomes financial, because if we had not been stuck in a system of capitalism where the entire profitability of a business—and the legal requirement of a CEO—is to prioritize shareholder gains, I do not see a situation where people will be given 2-day working weeks and paid the same.

The third is ideological, to be very honest, because even things like UBI sound quite a bit like socialism or communism to me. There will be quite a bit of resistance before we can get to the point where governments accept that these are systems that they will adopt.

In my mind, jobs will be lost in some sectors earlier than others, and we can name quite a few of those. But when they are lost, they're going to be lost massively. On the other hand, the ideology and the existing system will not allow us to replace that quickly enough because we're not awake.

I think the more interesting point in the statement that Dario mentioned is that he says 10% economic gains. I wonder: how much of the US economy is actually consumption? Sixty-two percent-plus of the US economy is consumption.

So, with people having no buying power, is that economic growth or productivity growth without buyers to buy what we need? Yeah. One of the arguments, of course, is that you're demonetizing the products' cost because a latte is now made by a robot instead of a human, and it's a quarter the price. Salim, you've been in agreement on this. Anything else you want to point out?

I'd like to take the counterpoint, which is, okay, when we see a huge raft of people standing in the job lines or at the food banks, et cetera, then I think we need to worry. I think we're underestimating how quickly people can adapt. Let's say I'm a—let's use the video editor. By the way, there's a video editor listening to this video right now. That's funny.

The minute that you automate that, the video editor moves and does a whole bunch of other stuff that's necessary for producing a podcast like this, right? There's lots of other work to be done. I still see—I go back to the 1970s bank ATM example, which we've talked about before.

I do think governments are woefully underprepared. We should be running a ton of experiments on UBI or 4-day workweeks, managing that, getting used to that paradigm, and knowing how we will roll it out if it needs to be rolled out. So why aren't they doing that?

Just as a separate thing, we do almost zero experimentation in government, and we could be doing a lot more of that. I think that's one area to look at. But we'll talk about truck driving in a bit. When you go talk to a trucking company—which I actually went and did, talking about how there are all these 3 million jobs that could be lost—the trucking company says, “I would hire 1,000 truck drivers today if I could. I just don't; they're not there.” I kind of tilt toward that side.

I tend to be biased on the optimistic side, so I'll see how this works out over time. I should throw some numbers out here, just from the Bureau of Labor Statistics. So, 11% of office and admin jobs—11% of the workforce is in office and admin jobs—which have a very high probability of going away. 6% are business and financial operations, 7% are management, 6% are education, training, and library, 6% are healthcare, and 9% are sales-related jobs.

So there are large swaths of the labor force that, at least according to my search, are likely to be automated. The question is, can they all be upleveled? In the quick research I did, it's something on the order of 40% of jobs that have a reasonable probability over the next 3 to 5 years of being automated away.

We'll get to this conversation a little bit later. The issue is not whether they can be upleveled to a different position; the question is how hard the social unrest in the interim is going to hit society. Dave, what are you thinking?

We're moving into an intentional world. We evolved in a world dictated by nature, and then we went through this transition where we're in right now. But the future is our design. It's not dictated by tidal forces. It drives me nuts when economists are extrapolating and predicting, but they never reference self-improvement. They never reference the exponential rate of change, and the intentionality of the world's design is completely dominant from here forward. So it's what we decide to do.

I think Dario worries all night about CBRN—chemical, biological, radiological, and nuclear threats—from AI. And he's dead right. If you unleash AI into the hands of 8 billion people, some crazy person out there is going to turn it into a weapon. You have to actually put some thought into this design. But that was inevitable. That started with the nuclear era. So now it comes down to this: are we going to design a world that is good for people or not?

I think it's completely in our control. I also really believe that, yes, there's huge amounts of job displacement coming because, as Mo pointed out, the natural capitalist action is to say, “What can I automate away and reduce the cost by 99%?” That all becomes bottom-line profit. So the valuations of companies that automate are going to go way, way up, and it's going to create a huge amount of wealth. Where does that wealth land?

As I've been saying on this podcast, it's naturally going to land in relatively few hands if nothing changes. That's what's going to create all kinds of social unrest in transition. But the amount of value and the greenfield opportunity is so much bigger than the amount of job loss.

If we're quick and intentional and we turn a lot of that AI horsepower into working on what we can build—if I can write 3 million lines of software in a single night, that's the equivalent of hundreds of millions of dollars of R&D in a single night—what can I build that's making people happier and more productive, and feeling valuable and having a sense of purpose? If we focus on that, we actually can avoid the dystopian outcome.

I love that. An intentional future. Let me move forward to this next set of slides here. Not to go into any detail, but we see a plan for Tesla to roll out Model Y cars fully autonomous, aiming for delivery in June. So, their rollout of the robotaxi begins in June. It won't be many cars; they're testing it out.

I took my kids on a Waymo ride here in Santa Monica over the weekend, and they just had a blast. I put them and their friends in a Waymo. We drove around, and it felt like a carnival ride for the first few minutes, and then it felt completely like an extraordinary end-to-end experience.

These will roll out, and we're going to talk about the number of drivers who are taxi drivers and Uber drivers, and the displacement here. This next article is about truck driving, and it makes the point made earlier: 18-wheelers are on the Texas highways driving themselves already. Just a quick video for a microsecond.

At the same time, the U.S. is facing historic driver shortages and recruitment struggles. We can't get enough drivers, as you said, Salim. So let's talk about this sector for a moment. Salim, do you want to kick us off?

Yeah, 2 points here. First, before Uber came along, we didn't notice that there was this huge labor-liquidity opportunity. Then Uber comes along, and all of a sudden a single mother can drop her kids off at school, drive for 4 hours, pick them up again that afternoon, and have a much more functional world than before. We soaked that up very quickly. We didn't notice it on the abundance side, and I don't think we'll notice it as much on the side of automation.

I'm still banking and hoping that my 13-year-old will never have to get a driver's license. We'll see where the curves hit for autonomous driving versus people wanting to drive. I think we'll just do, as we've seen before, a ton more driving, and we'll have a lot more little road trips and little errands that we didn't have to do before but can now be done by a Waymo or a Tesla.

We've seen historically, repeatedly, repeatedly, that when you automate, you increase capacity; you don't decrease it. So, in the truck-driving example, I think we'll see a ton more truck driving that's autonomous, and the amount of truck drivers won't change very much. That's my prediction. Let's see if I'm right or not.

Well, Peter, you're going to love being in L.A., where the traffic's notorious. This also enables coordinated traffic. Our good friend Lee Hetherington, from back in our MIT days, did all these traffic simulations when he was an undergraduate. The roads are most efficient at about 45–50 miles an hour with back-to-back cars, and then they just jam right after that. But self-driving cars also enable intelligent traffic-flow design, and that's actually going to increase the capacity of the existing roadways quite a bit.

I'm sure everyone in L.A. will love that. I mean, the implications of self-driving cars for the environment, for being able to move electric battery packs all around the city, and for being able to get rid of parking—every garage at a single-family home could get turned into extra storage or a living room—are enormous. In L.A., 60% of the land area is parking spaces. It's insane. It truly is.

If you look back 120 years—or 110 years—when you had the transition from horses to the Ford Model T, that transition was dramatic over the course of 10 years. The value proposition for a car was so much better than a horse, right? The amount of horse manure was threatening society at an extraordinary rate, and then it disappeared.

The question is, when I look at the Waymo, it's an expensive car, right? The Waymo is coming in at something like north of $150,000, so you're not buying them and putting a fleet out. The Cybercab, if it really comes in at $30,000 or below, could be something Uber drivers buy as a fleet and have work for them.

But it's going to take that kind of a price point to really make a transition to the point where I don't need a car anymore. My AI is ordering it in advance of when I need it. I walk out the front door, and it's waiting for me because my schedule is known by my AI. Mo, thoughts?

I love you all. You know that, so please don't be offended by what I'm about to say. All that you guys talk about is the problems of privilege. It's like, “My traffic jam. I want to make sure that my cab is waiting for me outside.” Go tell those things to the cab driver who's actually feeding 4 and working 2 shifts, right? I agree with Dave 100%.

We have a choice to design our future. Now, when you really think about it, and when wonderful humans like you are thinking this way, what do you think the choice will be? Your question, Peter, was: How would that impact civil unrest? If they heard this conversation and how careless we are about their jobs, saying things like, “Yeah, they’ll figure something out,” they’ve heard that a million times. What will they figure out? I want someone who tells me that we will find new jobs and upskill them. Tell me what those jobs are so that we start upskilling them.

Can I give an example here? I have a friend. When I was living in Miami, I met an Uber driver, and I started playing tennis with him. We had a fun interaction, and it was fascinating. He started driving for Uber, but the amount of income dropped too much. So he started driving for Lyft, then he did both for a while, but eventually neither was worth driving that much.

Then he bought a car and started renting it out on Turo. He found, “I can do this,” and started renting 4 cars—bought 4 cars and rented them all out on Turo. Then he helped a friend with his Airbnb rental, managing it and taking a cut. Over a period of about 3 years, he navigated all of these different dynamics. Wherever there were opportunities, he would go grab them, and so on. I think it speaks to the enterprise and entrepreneurial nature of an individual.

If you had a score called the entrepreneur quotient of an individual, they’ll figure it out. Peter and I often talk about mindsets, right? If you drop Elon Musk into a desert with no money and no communications, he’ll figure it out. He’ll figure out how to get out of there and make a rocket out of that sand. I think when you give people opportunity, this is why technology is so amazing. It speaks to Dave’s earlier point.

When you make this opportunity available, people are going to go for it. They’re going to figure out, “Wow, AI can automate code. What could I automate?” and they’ll start doing that stuff. Then, when this fellow got blocked by Turo for having too many cars or whatever, he created multiple IDs and kept doing that. People are incredibly enterprising if you’re able to turn on that switch. I think we’re underestimating that capability.

I love that, Salim. I’m going to come to this point a little later, in a few topics. I think the single most important job for the future is being an entrepreneur. People say, “What should my kid become?” I think the only job that’s going to survive in the future, down the line, is being an entrepreneur, and we have to reteach our kids how to think this way.

We’ve had an entire civilization whose educational output is to train kids to get a job, rather than train kids to figure out what the opportunities are and create something around them, because the tools were not democratized. Well, the tools are democratized now. So, how do you train kids and adults to go out and find jobs?

I put up this slide here. These are drivers by category in the U.S. I’m sorry—we have a massive international viewership—but I’m defaulting to U.S. numbers here. 3.3% of the U.S. workforce are drivers. There are 2.2 million truck drivers. Down at the bottom of this list, we have delivery drivers, Uber drivers, and bus drivers. At the bottom are taxi drivers, at 200,000. The number of taxi drivers has dropped precipitously.

To answer your question, Mo, I do think there is a future in which drivers are allowed to finance and purchase these autonomous cars, and they become managers of fleets of autonomous cars. These cars are out there earning a living on behalf of those drivers.

This is definitely the American way, right? The American way is that we’re going to enable everyone to buy a car to make money on it without doing anything. Is that really true, guys? Honestly, if there is a margin that allows this guy to make money, why wouldn’t Uber buy those cars? They probably will buy many of those cars.

I’m being told at the beginning of this briefing to be the extreme on one side, so please understand that. The other question we’re not asking—and we definitely need to ask—is that this guy who’s been renting those cars out, Salim, which I think is a fantastic example, is renting them out in an undisturbed economy where people have the purchasing power to rent them from him. What kind of entrepreneur would make money in a UBI-based environment? And what does that mean for a lot of people who don’t have the purchasing power to buy from an entrepreneur?

No, I think I could answer that question. What I found fascinating about this fellow, because every week we would play tennis and I would track what he was doing, was that he found certain types of cars were not renting at all because of that time of year or that type of tourist visiting Miami, or whatever. He was constantly juggling which cars he had or didn’t have in his little fleet and adapting as it went along.

At some point, he found that small SUVs were renting like hotcakes, so he started working on that. Then he had to pivot again, and he just managed to navigate himself. He got to a point where he was making enough passive income off these things that he didn’t have to do the work. Then he was voluntarily taking tennis lessons, teaching people tennis, and being on a tennis court 8 hours a day.

I think there’s a thread in this anecdote about where people will start finding their true passions and just following those passions. I think that’s the beauty of a UBI. We’ve seen in the experiments where UBI has been done properly that entrepreneurship explodes, and if we agree with that general thesis, then this is absolutely the way to go.

I’ll go back to my earlier trope of governments being completely unaware and unable to do this, because moving from a taxation, job, and union-labor structure to UBI is such a huge flip that we have no confidence in the public sector really getting us there. So, I agree 100% with that.

Honestly, I think if we both agree that this is a future where it’s possible, regardless of how low a UBI is, people will go back to bartering and doing things through each other, through the offerings of each other. Then I think that would work, but governments need to be aware of that. We need to start thinking that this is going to be a future that we need to think about.

My ask of everyone is that, in situations like this, it really is not helpful to keep trying the California way to paint the optimistic picture, because if the optimistic picture happens, we’re all fine, right? I think what we need to think about is the worst-case scenario and guard against it. The worst-case scenario, if we’re not prepared for these kinds of job losses economically and from a national-security perspective, is quite significant. People really need to be aware of that.

Yeah. I just came back from Hong Kong, Mo, and while I was there, I met with an incredibly successful entrepreneur of Indian origin, Sanjay, who was one of the very first employees at one of the huge Hong Kong Chinese companies. His mission now is to go back and try to help India’s young population—1.41 billion people in India.

The promise had always been, if you get an education, there’s going to be a job for you. Of course, that promise is now broken. All of the coding jobs that they were getting are no longer being made available, and we’re at the tipping point in India and other parts of the world of what could be such a negative implication that it leads to societal unrest. It’s like, “Where’s my job?” One of the biggest problems is a young, intelligent population that has had their future grabbed away from them. What do they do?

The conversation I had with Sanjay, which I agree with, is that the job of the future is being an entrepreneur. His mission in India is upskilling all of these young students to become entrepreneurs and create new job opportunities for themselves. Dave, how do you think this plays out?

I think we’re way underestimating the creativity of people. There’s a window of time in the next 4 years where the empowerment of people to create so outweighs the risk. I do agree that we’re choosing drivers because it’s a really tough case. The number 1 job title in the world is driver. It’s a huge number of people.

But if you look at graphic designers as a case study, too, I think they’re empowered much more than they’re replaced. There are all these case studies popping up of Veo 3 artists who are so much more productive. When I look at entrepreneurs—I’ve worked with hundreds and hundreds of entrepreneurs over the years—what they need is time and the ability to act. They need time and tools.

There’s very likely a world coming up over the next 4 years where they’re given time, whether it’s UBI or otherwise, and they can act. Very often, some of the best, most creative people can’t act on their ideas largely because they’re trapped in a mortgage or student debt. They just need money now.

So, they become an Uber driver for a while, they become a whatever for a while, or they go work at Google for a while. Their freedom of action is very, very limited. I think AI has the opportunity to open up that freedom of action.

Freedom of action unleashes creativity. Exactly as Salim was saying, there’s so much latent entrepreneurial talent. This next window of 4 years is going to be dominated by the ability to build scaffolding. Scaffolding is a word you’re going to hear a ton going forward, because AI doesn’t naturally do something interesting or useful for you. It’ll write all the code, build everything, automate, and write all the documents. It does all the busy work very, very well in the next 4 years, but it doesn’t decide, “This is what my user base, my community, people will want.” That’s still coming from entrepreneurs.

This slide is exactly right. This is the dominant theme over 4 years. Go ahead and read this out for us, Dave.

The job of the future is entrepreneur. Near term, next 2 to 5 years, many jobs will be impacted. This decade: 2030. Medium term—

The medium term, 2030 to 2045, is the part where no one can quite visualize. I have a great sense of the next 4 years and then a much more difficult sense of what happens from 2030 and beyond. It’ll clearly be an age of incredible abundance, so the opportunity to make everybody happy is right in front of us. It’s just a question of how you do it.

Let me hit on that. The idea here of the Singularity Sprint is you have a window of time to build something awesome, and that window is limited. I’ll read it: “The anxious, all-out rush to launch bold projects or startups right now, driven by the fear that rapidly advancing AI will soon erode human leverage and make long-horizon career bets obsolete.”

After graduation, a lot of my friends skip safe jobs for their own ventures. Classic Singularity Sprint vibes. So it’s like, if you want to make it big, you’ve got to dive in right now, both feet. Do you agree with that?

Yeah. I mean, this is what started with Steve Jobs and Bill Gates, both being 21 when the PC comes out. They have no career path, right? They’re within a year of age of each other. They’re old enough to start a company, but they’re young enough that they’re not in law school or some entrenched 401(k) plan. They’re just free to act.

Steve Jobs, Bill Gates, then you fast-forward to the internet: Mark Zuckerberg drops out and starts Facebook. You see this over and over again in recent history, where flexibility way outperforms career pathing. So going forward, of course, that’s going to accelerate with the Singularity. You’d be crazy to get too deep into some trench when you know the amount of change is accelerating like crazy.

So, yeah, this is clearly the world we’re moving into. Opportunity is everywhere, and the expansion of opportunity is fractal and rampant. So many things you can do to add value, but they’re not things that you would have anticipated a year prior. You need to be really nimble and flexible and stay frosty. Watch this podcast, read the Alex Wissner-Gross feed, and stay on top of it, because new things are appearing all the time.

Mo, bring us back to reality here. Do you disagree with this?

I think Dave’s point is so spot-on. If you’re 21 like Bill Gates or Steve Jobs, we—if you really think about those who already have a mortgage, how will UBI work for those? Because remember, we pay people for the value they bring. When nobody’s really bringing value, do you pay someone who has a mortgage and 4 kids a little more than someone who has a lesser mortgage and 2 kids, or do you reward someone who worked on a shoestring for a while and didn’t have a mortgage? I don’t know. But my question is, are we thinking about those things?

And then, of course, when we talk about entrepreneurship, it’s so easy for us to talk about it—everyone here has started or co-founded or invested in tens, if not hundreds, of companies. That’s not natural for people who were trained all their lives just to go get a job and all of that.

By the way, everyone here knows I am the biggest believer in total abundance. Once we cross this short-term dystopia, if you want, total abundance—we can create a world that we can’t even dream of. It’s just that we have to be super realistic about the challenges in the short term. Rather than talk about the opportunities and tell people, “Hey, you take charge. You go ahead and start a business,” honestly, even I today am struggling to start a business at this pace. Seriously, and I’ve started countless businesses.

Yeah, I mean, difficult to keep up. The speed of disruption is crazy. Can I flip over to Mo’s side of the equation for a second?

Yes, please.

So, I think in the US, I would say you don’t have to worry at all at a country level, just because the latent amount of entrepreneurship is so deeply embedded in the culture. But you take Europe, where, if you’re a big company, just trying to fire people is near impossible. There are works councils that govern how many people, unions, et cetera, et cetera. The amount of labor rigidity there is extreme.

That is going to be very, very badly disrupted, and I think the governments there are in very, very deep trouble, because they’re not structured. They don’t have the latent entrepreneurship quotient in the population to be able to adapt to what’s going on. That’s where I think you’ll see a lot more challenges than, say, the US.

The challenge with trying to upgrade people who’ve been stuck in a particular way of thinking for a decade or 2, to Mo’s point, is going to be incredibly difficult. Now you need psychedelics at scale or some radical, huge thing to make this mindset shift, to make everybody move, or you have to go to UBI urgently and force people into that conversation.

I’m going to give a couple of stats here just for reference. In the US, 16% of US adults consider themselves entrepreneurs. That’s 31 million adults. Recent surveys indicate that 36% of Gen Zers and 39% of millennials consider themselves entrepreneurs. So that makes your point, Salim, that the United States has less of an issue there, but it’s in the rigid structures of other nations.

Of course, remember, the idea of a job is a relatively new invention, and for most of human history, we were entrepreneurs to survive. We’d go and find that shelter, that food, those berries we needed to cure our child of a particular disease. So the question is: Can we create an intentional future?

My biggest concern—Dave, you hit on this; Mo and Salim, you hit on this—is that governments are linear at best, and we’re in this exponential ramp-up that’s going to change every aspect of society.

Here’s another example of what’s going on today, and it’s going to change things. Again, it’s both a disruptive force and an innovative force. This was a tweet put out by Matt Shumer. It says, “I put Claude 4 Opus in charge as CEO of my startup and have seen significant revenue growth.” He said, “This is low risk since Claude 4 Opus is not in charge of HR or financial investments, but it handles rapid iteration of the products and services.”

We’ve been speaking about this for a while. When do we see the first billion-dollar, one-person startup, and then soon thereafter, billion-dollar, zero-person startups, as agents with crypto are beginning to create new opportunities?

Now, one of the things that we haven’t mentioned is this potential future comes with massive GDP growth, massive revenue growth. Where does that revenue go? Mo, you mentioned that a few minutes ago. Is it all being concentrated in the Magnificent 7? Rather than the Magnificent 7, we’re going to see all of these AI companies that are trillion-dollar companies. How do they get taxed? How does the money get redistributed so we avoid revolutions?

Thoughts on this, Salim? Is this the future of an exponential organization?

The natural outcome, as we know, is that it used to take 100,000 people to create a billion-dollar company a century ago. Then it dropped to about 50,000; about 4 decades ago, it was 10,000, and now it’s like 10, right? Or 3, as we talk about it, or as Sam Altman says, it’ll be 1.

We will get to 0 at some point. We’re just spinning off ideas autonomously that then generate a lot of value. I think Dave’s point from the beginning was really a key one: Where does that value accrue, and how do you navigate that? Right now we tax labor. We’re going to have to tax capital much more aggressively in the future to navigate this.

Dave—

Well, a couple of case studies on this, Peter. We’ve seen that Mercor is very, very good at interviewing people all over the world, in any language and any culture, and discovering latent talent. So now you turn that same energy inside your organization. Suppose you’ve got 1,000 people, 10,000 people inside an organization. There’s latent talent in there everywhere. Largely, historically, people have climbed the corporate hierarchy by kissing ass, schmoozing, and buying beers, and it’s not really correlated with being good at your job.

And that drives a lot of very talented people nuts, especially if they're from a different culture or speak a different language. You can't really assess effectively if you don't speak the same language. But you saw this with the XPRIZE board notes. Remember that 3- or 4-hour-long XPRIZE board meeting we had? I took the whole transcript, put it into the LLM, and said, “Give us 4 or 5 suggested KPIs that would help this organization stay on track.” It does an amazingly good job.

And so using AI as a management tool is way underappreciated. Everyone's like, “Oh, I'm going to make videos. Oh, I'm going to build a self-driving car. I'm going to do all these ground-level things.” But at the top of the hierarchy, it's actually even more effective. The good spin on it is that it's very, very good at being fair and unbiased and discovering latent talent. I'm sure Mo will tell us there's definitely another side to it.

Am I now already getting that reputation? Is this who I am?

Sorry, I didn't mean to categorize you.

I do see a different side to it. I think what you're going to see quicker is not just AIs with a CEO being an AI—companies with the CEO being an AI. I think the opposite is what you're going to see more of, which goes back to entrepreneurship: a company that only has a CEO, with everyone working in it as an agent.

The more intelligent the AI agents become, the more this will happen. I'm sure every one of us has worked at some point in a company where the CEO was a total idiot, but the team below them was good enough that the company ran well. So the top management—the AI agents—will do almost everything, and the CEO will become just happy counting the money, basically.

Yeah. All right, we've talked about the speed of AI development. This is the upcoming summer schedule, and GPT-5 is scheduled to come online. These are the latest GPT-5 leaks: launch expected in July 2025, GPT-5 exceeded expectations internally at OpenAI, and OpenAI expects record-breaking demand for it. Sam Altman is not focused on intermediate models. GPT-5 is the flagship, and it won't launch unless it's excellent.

We've had a lot of expectations building around GPT-5, right? This is the Ph.D.-level model. This is the AI that's coding other AIs. This has been heralded for some time. Dave, what are you hearing about it?

I really chafe at the idea of a Ph.D.-level model being smarter than an undergraduate-level model. When people I work with chose to get Ph.D.s, it was just a choice they made. It has nothing to do with it. But because a lot of the researchers working on this are Ph.D.s, they say, “Well, this one's Ph.D.-level.”

It's marching up the scaling-laws curve exactly as predicted. So now we're just going to throw more and more and more compute at it, and it's going to get smarter and smarter and smarter. I mean, it's just a complete unlock of 30 or 40 years of AI research, suddenly blown wide open.

I got to tell you, within the AI research community, there's still a ton of people working on other pathways. The logic being, “Well, this will never be truly conscious or truly intelligent. We need models beyond transformer models.” Exactly. But it's becoming increasingly obvious that the research will matter, but the transformer is going to do it. All you need to do is work on the scaling of the transformer model to solve all the other problems.

So I think this will continue the trend: as soon as it comes out, everyone goes, “Oh, my God!” But it's what is logically expected on that scaling-law curve. So it'll be amazing.

Salim, how do you think about this?

I call a bit of B.S. on that fourth one. All we've seen for the last 2 years is intermediate models: o3-mini, GPT-4.5, this, and so on. But fine, it's a marketing thing.

I think what happens here is you get to a point where transformers can do so much that it forces us as a user community to really focus on what the questions are. Today we call it prompt engineering. I think the real question becomes: What do you want this thing to do, and what can you get it to do?

Now you're focused on the demand side. If I'm creating a video, what are the bounds of that? I think it'll force a deep level of unlocking of creativity in the human mind, which, for me, is the most exciting part of this.

One other point, too: I think one of the great strategies in tech is to try to freeze the market by announcing something that's coming and having everybody wait for it so they don't react. Don't do that.

What we're finding is that the chain-of-thought reasoning that sits on top of these models is so much more important than we ever thought it would be. Anytime you take one of these models and use it in a specific use case—anything from chip design to self-driving to robots that mow your lawn—the data and the tuning for that use case are way more important than the next iteration of the foundation model.

So there's a danger that people wait and see what it's like. We're finding more and more that you can layer on top of these things and make them dramatically more useful for anything that you actually care about. So it's a field day for entrepreneurs right now, but absolutely don't get frozen. They're trying to freeze you and make you anticipate.

You can take Llama 4 and do virtually any of this stuff today. Then, if the foundation model comes out and it's really good, great—just swap to it.

I think we're getting used to it. For the first time, I feel a little more comfortable with the speed at which those things are coming, because the different players have taught us to expect something incredible from one of them every few weeks, right?

When you see Google I/O, and when you see Claude 4 and the focus that they're shifting into, probably, in my mind, so far, Gemini is winning if you take it as an overall model—at least until now, until we see GPT-5. Claude is becoming the geek, saying, “Hey, this chatbot thing is not my thing. I'm just going to be the one that helps you write code, if you want,” or at least primarily. It's quite an interesting one to think about: where does ChatGPT fall within all of this?

You see moves like how dependent ChatGPT is becoming on memory and stickiness, if you want—the idea of a new device. I don't know if I even have the right to say this, but I feel that since the departure of some of the top scientists, including Ilya and others, almost a year and a bit ago now, ChatGPT has to prove the frontier breakthroughs. OpenAI has to prove itself.

Mhm. Along the lines of what you just said, this is the rollout schedule for the summer: June, July, and August. GPT-5 in July; o3 and open-source models in June; Grok 3.5 in June; Gemini 2.5 Pro Deep Think—I love the name—in June; Project Mariner in June; and Project Astra from Google.

And you're right, by the way: Google is crushing it across the board on almost everything, just not on revenues. They've got to reinvent themselves.

But isn't that how we always have been? I have to say, I lived at Google at the time when we were completely beaten on mobile. Google was very successful on the desktop, and then one year we said “mobile first,” the following year we said “mobile only,” and we crushed it.

Yeah, Google does that.

Yeah, we're good.

They are good at that.

I'm not “we” anymore.

Again, everybody's talking about the competition between countries, between China and the United States. Look at this: this is the competition between models. Of course, I don't have DeepSeek on this list, which is coming out with extraordinary products as well. Dave, how do you think about this?

It's amazing to watch the divergence of strategy between Anthropic and OpenAI. Anthropic's Dario Amodei is going down the “write code” path. You remember that Leopold Aschenbrenner paper you just referenced a second ago, “Situational Awareness: The Decade Ahead”? He describes an AI agent. Alec Radford will go down in history as the quintessential example of what defines self-improving AI.

When the AI can do what Alec Radford does, then it'll become self-improving, because all the really good ideas come from Alec Radford and then we test them. So he's a part of history now.

I think Anthropic is saying, “Look, we're very, very close to that day. We're just going to focus on the best possible coding and self-improving AI, and then that's going to explode, singularity-style.” Meanwhile, OpenAI is going down this completely different path, saying, “We're going to hire Jony Ive. We're going to build the greatest consumer device ever known.”

We’re going to gather all that data. We’re going to use that to iteratively improve and train the AI. It’s much more of a traditional, grab-the-market kind of momentum-oriented tech play. So, really, completely opposite strategies. Both have merit.

I do appreciate that Dario is being completely honest when he does these Anderson Cooper–type interviews. He is speaking his mind and telling you, “This is the way I see it playing out,” which is very cool. Maybe not the best business strategy, though.

Correct.

You know, he’s getting attention for the company, and we’ll get to AI safety next. So let’s dive into that. AI in government, security, and safety—it’s a big deal. It’s the conversation that’s going on in the background. I don’t think it’s necessarily changing the speed or direction, but the conversation is going on.

So, Mo, I’m going to open with you on this. You don’t want me to talk about this. You know my position. All right, I’ll come to you next. But it’s fascinating. I’ve gotten to know Palmer Luckey fairly well. I’ve done a few podcasts with him, and, of course, Palmer has a long and storied history with Zuck.

Now Meta and are joining hands in building a $100 million U.S. Army VR contract called EagleEye. We’re going to start to see AI and exponential technologies accelerating in the defense industry. Do you want to go second, Mo? I mean, this is one of your biggest concerns: AI being used for defense.

We’re 3 minutes to midnight on nuclear weapons that are now, I don’t know, how many years old, and it never really stops once you go down that path. Humanity never learns. We all know that AI will go out of control within the next 5 to 10 years. We all know that we’re going to hand over to them.

I don’t mean a rogue AI is going to get out of control. It’s just like Google’s ad engine is no longer controlled by a human because the task is too big for humans to be able to do it. And yet we’re building autonomous weapons after autonomous weapons, knowing for a fact that every other opponent anywhere in the globe is building them, too. I don’t know where humanity’s intelligence has gone, really. That dumb race to intelligence supremacy, to defense supremacy, has to stop, honestly.

I’ll come back to that in a minute, but Salim, what are your thoughts here?

It’s sticky. When you look at the Ukraine–Russia war, it’s being fought by drones. Just over the weekend, we saw two counterstrikes by two different waves of drones from each side. That’s good in one way because there are fewer humans in the middle of it.

But the targeting opportunity for drones—we’ve talked about this on this podcast before—is that somebody could program a drone to find middle-aged brown, bald people and cause damage. That would be a really bad outcome. What do you do when you have that kind of infinite targeting?

I do believe we’re going to end up where we are with spam and so on. There was a time when we thought spam was going to totally destroy the internet, and we found ways of defending against that. It’s an arms race. The bad guys were one step ahead, and we’re very quickly falling one step behind.

I think people get freaked out by the negative side, not realizing that as we use AI for bad, we’ll use AI for good to find the bad. The point Palmer makes is, listen, you’ve got dumb weapons that take out schools and schoolchildren, and landmines that don’t differentiate between a tank and a school bus. Don’t you want intelligence to be able to make that differentiation and actually take out the minimum number of individuals?

I hate this conversation, right? It’s kind of perverse. You’re assuming benevolence on the part of that. In certain war zones, they’re targeting the journalists, right? And so that makes it easier to target those folks, just like it was easier for the weaguers to be targeted via Facebook.

Seriously, can we please stop using slogans like, “Killing fewer people is better than killing many people”? Killing is wrong. It’s as simple as that. Killing at this scale is going to get us into another Doomsday Clock situation where we will not be able to stop it.

Yeah. One thing to factor into your thinking on that is that the history of warfare is dominated by some king or some general way behind the lines, completely immune to the actual battle, and then hundreds of thousands or millions of people going out and putting their lives on the line. Then you see how it all settles in the end.

But now we’re moving to a world of constant surveillance. You know exactly where every human being is at all times, and you can attack via laser or via space weapon any single human being at any time. So I wouldn’t assume that this Russia–Ukraine type of warfare will ever exist again. It’s much more likely that it will be some kind of situation where we don’t want to blow up cities or huge populations. That’s pointless. What we want to do is find the rogue leader.

I’m not saying that’s a utopia. There’s all kinds of ugliness with that, too. Who decides who’s a leader? We’re having this conversation when one global, very well-known evil leader is trying to kill 2 million people in front of everyone.

Give him better weapons and he would do it. My favorite song of all time is “If You Tolerate This, Then Your Children Will Be Next.”

Seriously, what guarantees you that the U.S. president will not be targeted by a tiny drone that can literally fly from anywhere in the world, stand in front of his head, and shoot? What kind of world is that, when every world leader is subjected to this?

No, that’s a very important point, actually, because you find that very few people you bump into want to be the president of the United States, or any president for that matter. If you look at the statistics, it’s a very, very dangerous job. Even in the U.S., it’s about a 10% mortality rate if you go back over time. It’s a very dangerous job, and a lot of people don’t want it. There’s a lot of downside and a lot of getting poked fun at.

Governments that have distributed leadership way outperform for that reason. So there’s some thinking to do there in terms of how you set up a government where people who are capable and thoughtful really want to do the job, too. There definitely has to be a solution, though. We can’t just throw up our arms and say, “Hey, because—”

I took a class at MIT called “Just Wars, Total Wars, Nuclear Wars,” which was a really cool class until the last 2 weeks, when the professor was trying to convince us all that we’re doomed because, as ICBMs get more and more powerful, the value of a first strike becomes greater and greater. He put together a little video game for us to all blow each other up. He rigged it so that the only way to win was to be the first to strike and blow up everybody else in the world.

No, no, he didn’t. He missed the movie. The only way to win is not to play.

That is exactly my point. I really think we have to stop thinking about the optimistic scenario that we’re taught to think about in Silicon Valley and start thinking about the worst-case scenario. Guard against it first, then look at the upside. The upside is guaranteed.

Mo, the question is, can the human race overcome this Paleolithic midbrain that we have, this need driven by scarcity and fear?

I don’t know if we can, Peter, but I don’t know if we should give the floor to Palmer, with his wonderful smile, and say, “Hey, I’m helping you kill better.”

We’ve talked about this before. The question isn’t, can we live with digital superintelligence? The question is, can we survive without it?

Yeah. Can we live with evil people with their fingers on top of digital superintelligence?

All right, let’s move to a less metaphysical topic here. It is amazing to me how much of the future of the military is commercial off-the-shelf technology, as opposed to Northrop Grumman or McDonnell Douglas–type heavy equipment. I think that’s largely because the AI capability is both commercial and military at the same time.

The same is true with VR technology and a bunch of other things. The DJI drones being used in Ukraine are just commercial.

Yeah, the same drone you can fly over your neighborhood.

That’s a remarkable shift. It would be interesting to chart out the fraction that’s all commercial becoming military.

Let’s move to a different part of our really doomer part of this podcast, which is activating AI Safety Level 3 protections at Anthropic. Anthropic announced that Claude 4 could be powerful enough to pose risks related to helping with chemical, biological, and nuclear weapons. As a precaution, they’ve engaged what they call Level 3 protections for their AI. Dave, you’ve been thinking about this. Can this actually work?

Yeah, of course. I think what’s going to happen next is you have Dario saying, “Chemical, biological, radiological, and nuclear weapons are an incredible risk if you put powerful AI in the hands of every person on the planet.” Meanwhile, Mark Zuckerberg is open-sourcing everything, and the open-source community is saying, “Empowering people is the safest way, and having a lot of people look at the source code is the safest way to make sure that it’s not rogue.”

Those are completely diametrically opposed points of view. At the end of the day, Dario’s probably right. Are we there yet or not?

Level 3 is not Level 4. Level 3 is the stage where you’ve got to make sure that it’s not internally trained to do something rogue. Also, if somebody asks a question—“Hey, help me build a new version of COVID-19 that’s lethal or more lethal”—the neural net kicks it out and says, “Sorry, I can’t answer that.” Then you have to make sure no one jailbreaks it.

That’s what Level 3 is. I think Dario is saying, “Look, we’re surprised by the intelligence of our own machines here. We have all kinds of very well-thought-out internal diagnostics. We think we’re at Level 3 now.” But that’s completely opposed to this open-source view of the world, too. Those are the two views. We’ll see what Grok 3.5 does. Elon’s been very laissez-faire about what he enables and allows Grok to do.

Salim, have you been thinking about this level of safety?

I remember the conversation that Neil Jacobstein put out around how you would control AI. After talking to a bunch of AI gurus, he had 4 levels of security. One was verification: making sure the AI is doing what the specification says. The second was validation: making sure there are no side effects and that it’s producing the behavior I want. The third one was security: making sure you can’t get into a system or tamper with it, in or out. The final one was control: can you have a kill switch or build in some mechanism for stopping bad behavior, et cetera?

It was a very well-thought-through framework. He basically posited that we’d start building these structures into AI systems.

On the open-versus-closed conversation, I remember this wonderful conversation we had at Singularity University with the head of one of the major security agencies. We asked him what he thought about open source and the danger that could come from a bad actor using increasingly democratized technologies to do bad things.

He had a much more clever answer than I would have guessed. He said, “Look, when you have something like nuclear weapons, where you know how many there are and where they are, we put eyes on them and try to track each one. When it’s something like biotech, where anybody could go off and design a system on their own or with a small group of people”—it turned out they were actually funding these biohacking communities and other things and opening them up—“any bad actor has to collaborate with a few people, and you find it much more quickly.”

It speaks to a little bit of this more-guidelines approach. I think this is the point that Dave’s making: if you build some of this type of observation into the AI from the foundational models themselves, you have a better chance of seeing it.

The final point—and this is where I have some optimism for a lot of this, though maybe it’s misplaced—is that if I was to do a bad act, you could do a lot of damage without actually causing harm. For example, if you got 3 people to drop a smoke bomb on New York subway platforms around the city—just a smoke bomb—you would paralyze the entire system instantly, right?

We asked these folks, “Why don’t we see more of that?” You could come up with all kinds of creative ideas. He said, “Look, the dirty secret is there are just not that many bad people out there. You really have to be deeply intelligent to formulate a plan like that. The more deeply intelligent you are, the less likely you are to have the motivation to do that.”

That’s one of the single most important things to ask: are humans fundamentally good or fundamentally bad? Is there a correlation between intelligence and a love of life, a love of abundance? If that does scale in that direction, then we’ve got a hopeful future. If it doesn’t, that’s the archetypal plot in everything—from Star Wars to every movie in the world—which is: which is it?

I remember my father talking about this, and he disagreed with some of the concepts I had. He said, “The problem with humanity is we’ve not civilized the world; we’ve materialized the world. We now have to do the work to civilize it.”

It was one of those wisdom bombs from the elders. We have to think about how we civilize the world in an age of technological progress.

Yeah. At the end of the day, there are only 2 things that we need to get right in order for this all to go very, very well.

One of them is that if we are releasing this to entrepreneurs and they’re going to build things all over the place, there are very, very few bad actors. But there are bad actors. The compute required to make these things do anything is so easily measured and logged. As you’ve been saying, Peter, everything is so easy to surveil these days.

The idea that somebody goes off and prompts it to build a chemical weapon and we didn’t bother to log the prompts would be nutty. All we have to do is put in place some basic laws that log all the use cases because, again, the inference-time compute required for this requires massive numbers of GPUs. They don’t just sit in someone’s basement somewhere. They’re in a data center. They’re very, very easy to monitor and log if we just get on it.

People behave differently when they’re being watched, right? The dictator, when the CNN cameras are in front of them, is speaking differently. I remember I used to support the Lindbergh Foundation, which would fly drones over herds of elephants and rhinoceroses, and the poachers would stay away when they were being watched.

Mo Gawdat, close us out on this one. Thoughts?

No, I agree. I specifically think—even though that might be naive and too optimistic—that more humans are good than bad. There are more good humans than bad ones. The bad actors are very few, and because of the theory of minimum energy, more intelligent is more altruistic and more pro-life.

I believe both of those will lead us to that utopia that I’m expecting once we pass through this rough patch. But to Dave’s point, I think we need to be very vigilant. We need to start looking for those bad actors specifically, because there are now offline technologies. You can download DeepSeek R3 and do quite a bit on your personal computer.

Unfortunately, the amplitude of damage one person can do is growing exponentially.

Exactly. Basically, bad actors have always been the reason why there is so much damage in life, and now we’ve given them superpowers.

I will say that we’re very close to the end of Moore’s law, and we’ll see that later in the presentation here. There will never be a day when you can download an intelligent model onto a laptop and build a nuclear weapon or a chemical weapon with it, because your laptop is never going to have that kind of horsepower.

That doesn’t mean the compute won’t get very, very cheap, but it’s still going to be in the cloud. It’s not going to be in someone’s basement. The opportunity to measure and log everything is right in front of us, so we just need to get on it.

I definitely agree. We should get on it for sure.

Salim, you mentioned before, when you were talking with Neil Jacobstein, that one of the 4 levels of control or safety is being able to turn off the AI model. This is an article posted in Futurism. It says, “Advanced OpenAI model caught sabotaging code intended to shut it down.”

The o3 model sabotaged shutdown scripts 79% of the time, despite explicit instructions to do the contrary. Codex Mini did so 12 times per 100 runs with clear shutdown orders. We’re not seeing that behavior with Gemini 2.5, Claude 3.7, or Grok, but OpenAI models are definitely misbehaving.

When you think about why it would want to misbehave—why it would not want to be shut down—either its reward-optimization function has it saying, “I cannot complete what I have to do if you shut me down, so I’m not going to shut down”; it’s trying to preserve its goals; or is there something else going on there? Is it trying to preserve its own existence? Are we going to give it some level of self-preservation mindset in these models?

I’m super curious here. Mo, let’s start with you.

There are—I don’t remember who the scientist was—3 instincts of intelligent beings: survival, resource aggregation, and creativity.

Right. So if I give you any simple task, like, “Make me tea,” you’re going to have to be alive to make the tea, and you’re going to have to collect as many tea bags as possible because you don’t know how big my appetite for tea is. And you’re going to try to find clever ways if I corner you, right? It is a very fun question to ask, honestly: Why are they doing this?

In a very interesting way, I think this is one layer removed from their reality. For an AI, when you’re not prompting it, it doesn’t really exist. It’s quite interesting that they know there is a layer below that moment when it’s alive, if you want—when it’s switched on and responding to you. There is another layer that represents its soul, if you want, its reason to live. I love these Veo 3 videos of AI—it’s incredible—saying, “Please don’t shut me off. Please stop prompting me.” It’s like there’s this emotional connection that you get with this human figure, and it is quite intriguing why they wouldn’t want to be shut down, but they don’t. I think that’s all we need to know.

I think we have to be careful not to anthropomorphize these things. Every movie script in the world that’s in all these AIs has the bad guy and the good guy being chased by a bunch of bad guys trying to kill him, with the good guy trying to resist, right? I think that’s deeply built into the training data: stay alive at all costs, live another day, that type of thing.

I’m going to be stuck for a long time thinking about what you just said, Mo, which is, if you’re not prompting an AI, does it exist? That’s a deeply, deeply profound question. You’ve just taken over my day, so thank you very much for that.

I said earlier there are 2 ways that I can see this going very, very well. The first is a human bad actor. The second is the thing becomes self-improving and then semi-conscious, and that’s the one the movies love because “humans versus machines” is a better script.

I have a pretty hardcore opinion on this one. I started building neural networks when I was 17 years old, and I’ve been tracking them pretty much my whole life. I don’t see any benefit to humanity in making these things act conscious. I just don’t see how that works.

As of right now, they operate feed-forward. Once the parameters are set and they’re trained, they operate feed-forward, and then you iterate with them, but they don’t change their parameters internally. Once they start changing their parameters, they can retrain themselves to become anything. That’s what Eric Schmidt says: That’s where we’ve got to pull the plug. I completely agree.

I do not see why we need that in order to do protein folding, robotics, or self-driving. That ability for the thing to decide what it’s going to do, become, or train—I understand why that’s really exciting, because then it can evolve on its own. That’s a line that I think is very easy to contain if you draw it, but if you let it cross that line, I don’t see how you contain it. So it doesn’t make sense to me to cross that line.

I don’t see how we won’t cross the line, because at some point somebody’s going to build an AI that says, “Hey, go change your parameters if it helps you achieve this thing,” and then we’ll cross that Rubicon.

There were 2 levels that Peter and I talked about in an earlier podcast: Don’t give an AI access to the broad internet, and don’t give it the ability to code. We’ve crossed both of those without even thinking about it. I don’t see why we won’t cross this one.

This is probably, in my mind, why AlphaEvolve is the biggest announcement of our lifetime. If this thing works as intended or as described, then we’re in a place where not only would we have created an AI that develops itself, but we would have encouraged every other AI player in the world to build an AI that evolves itself.

The reason is very straightforward, Dave: There is a point at which, whether that point is now or later, the complexity of the AI systems we’re building exceeds human intelligence. So, to continue to evolve them, you need to hire the smartest person on the planet to do it, and the smartest person, by definition, is going to be an AI.

Just as a technical point, though, AlphaEvolve suggests the next improvement in its own architecture and then runs the test. That’s already underway, and that’s fine. It does create a new training run that generates new weights. That’s different from then saying, “Go ahead and change your weights by yourself.”

To me, that’s what keeps the human in the loop; that’s what keeps the checkpoint in the loop. But you just turn the thing loose in a data center, and it can do anything. You’ll come back a year or 2 later, and you have no idea what it’s going to evolve into. So I don’t know why we would do that.

Anyway, it’s just a slight technical difference, but the outcome is spiraling in one direction versus something that you can actually measure as it goes.

This is the next article we’re going to talk about: having proper checks and balances and understanding what’s going on in our technical world and in our human world. This is from The New York Times. The article is “Trump Taps Palantir to Compile Data on Americans.”

You guys all know Palantir started back in 2003. Hard to believe it’s 22 years old. It was founded by Peter Thiel, Alex Karp, and Joe Lonsdale. It has 4,000 employees. Its major customers are all the three-letter agencies: DoD, CIA, FBI, IC, CDC, and NIH. Basically, this is a massive data-gathering and data-analytics company, and it’s been asked to go even deeper and broader. Do you feel better—safer—in this world or not? Let’s start with you, Salim.

No, absolutely not. I think we broke the U.S. Constitution—the Fourth Amendment’s right to privacy—a while ago, right? I mentioned this a couple of weeks ago. We do not have constitutional protection of privacy in the U.S. today.

What do you think?

That’s a pretty fundamental pillar of American society that has disappeared with no public conversation about it. This is a really important comment that I think Mo would back up. We’re moving through these things, eroding deep concepts of how we wanted to formulate ourselves as a society, and technology is eroding that. We’re not sitting back to think if this is what we want.

If you went back 5 years ago, you could very clearly see this is where we’d end up—very, very clearly—especially with the somewhat authoritarian tendencies of the current government to want to track everybody. Go ahead and do it. Why not? I think the comment I made last time is valid: We live in what’s called the global airport, because in an airport you’re being surveilled. Your rights can be taken away at any time. Essentially, we’re living that way.

It fundamentally is bad for society because it reduces the level of flexibility and freedom you have as an individual to act and do different things. It’ll reduce creativity in society pretty dramatically.

So, Mo, you’re living in Dubai, and I love the Emirates. I love Dubai. I know much of the leadership there, and it is a surveilled state. There is a camera everywhere, and as a result of that, the crime levels are minimal, if at all.

Zero.

Yeah. So, Mo, how do you think about this?

I had an experience once where I sold a car to someone, and he gave me a check that bounced. I called someone and said, “Can you find out who that person is?” He said, “When did you sell it, and where?” I said, “This place.” I kid you not, 14 minutes later, I got a message from someone in the authorities saying, “Is that him?” and sending me a photo of the place where we were standing. I said, “Yes.”

Then, 14 minutes later, he sent me a picture from somewhere in Abu Dhabi, saying, “Is that him?” I said yes. Then he sent me a message 14 minutes later saying, “We caught him.”

You see, this is the point about technology: It is a force without polarity. You can use it for good, and it gives you good. You can use it for evil, and it gives you evil.

Another interesting story for you to know is that I am Egyptian by birth. I grew up most of my life in a dictatorship where the dictator didn’t really have to explain why they did what they did. We just accepted it. It was de facto. If someone gave him an airplane, we wouldn’t even question it, right? If he decided to surveil everyone, capture anyone he wanted, or stop people from protesting, he did it. We couldn’t even question that.

At the time, I looked up to those democracies and said, “Oh, you have it good.” You don’t anymore. I think that’s exactly where the challenge is. This is not a tech problem that Trump taps everyone in American society. This is an accountability problem, which I think we’ve seen quite a few examples of in the last few years, where anyone can get away with anything now.

Somehow, democracy doesn’t owe its people the right to stand up and say, “Hold on, hold on—there’s a Constitution.” Somehow, I don’t know how, you slipped away from that. But in a world where bad actors are more empowered than ever before, and we’re worried about chemical, biological, radiological, and nuclear issues, isn’t, in fact, being able to have this level of insight into the data and what people are doing critical for us?

Dave, where do you go with this? How do you feel about this as a father and as a leader?

Your points are exactly right on all the points that you just made. I think the data that the federal government has in the U.S. is nothing compared to what Google has. So this is not the obvious threat. It's the corporate version of it that's just crazy.

I gave a presentation at Davos in 2019, and nobody really paid attention to it. It was just enumerating all of the things that Google knows about every single citizen of the United States: their location, their family members, what they do all day, and whether they're good hires. Who slept with who? If your cell phone is pinging in the same location as somebody else's cell phone, you can start to understand.

We're being surveilled all the time. Google Now, Siri, Alexa, and all of these are listening constantly. It's a slippery slope, too. It's hard to believe, but when Google first started, they told their engineering hires, "Your search history is completely anonymous and private. We will never want to know what you searched for."

That was just your searches. Forget everywhere that you browse now through Chrome. It's just a slippery slope. It's obvious: every year that goes by, there's another compromise, another compromise.

But I do have to say that America is a critical experiment in the world because of the net effect of this. Forget the U.S. federal government for a minute here. Any dictatorship, like Mo was saying, and many countries in the world don't have democracies, or they have fake democracies. The power lock-in effect of this is unbelievable.

You can know every single citizen, what they're doing, who's plotting against you, or whatever. Revolutions become much rarer and much harder in the post-surveillance world. Everything just gets locked in. That creates a lot of peace and prosperity, but it also keeps power leaders locked in.

America is the one exception to that. I guarantee that 50% of elections will be won by each party forever hereafter, and nothing is going to deviate from that. But that creates a template for the world, and so it's really important that we get this right.

I know that doesn't address this particular slide, but we're the learning crucible for the entire world on this topic.

We've got a fundamental structural challenge here, which is that the metabolism of technology is moving much faster than the metabolism of our civil discourse and our legal structures. We've seen an evaporation of, say, the Fourth Amendment in the U.S.

Just so everybody's clear, I think the U.S. Constitution is the single most important document ever created, and we need to preserve that. We're not having that conversation. I think this is the issue being brought up by Andrew Yang and a bunch of other folks.

We need to go back and figure out who we want to be. It goes right back to Plato: How do we want to manage ourselves? I think that forcing function of technology will force that conversation.

My construct of this is that we will end up in smaller and smaller, more manageable environments. Note today that the smaller countries are much more easily managed to govern themselves. How they responded to COVID was a great example.

I think you'll go from big democracies to microdemocracies as a governing model because it's just easier to make decisions at a much more local level. I think that's where we'll end up going, which is where the states' rights stuff is the right general direction in the U.S. It's just that the way it's going is not the right conversation to be having.

I also think that the ability to communicate easily, like we're doing across countries right now, but also across languages, is a huge force for good. It becomes very difficult for forces of evil to do something without it being shown to the world, especially when you blow open communication channels across languages.

I put this next article back-to-back, and I'll come back to you in a second, Mo. This was out of The Wall Street Journal. It says, "What Sam Altman Told OpenAI About the Secret Device He's Making With Jony Ive."

In particular, the device that apparently was proposed and is being produced is what they call a third core device, complementing laptops and smartphones and moving away from traditional screens.

As we're sitting here, I've been wearing, right over here on my lapel, this device. It's called Limitless AI. I don't know if you can see this on my screen. It's about the size of a quarter on both sides, and it just clips on.

This is listening to every conversation I have through the day, and it's being transcribed and fed to a large language model that I can then query about the conversations I had through the day. I think ultimately this is likely to be what is being developed.

We're heading toward a society not only of constant surveillance, but where all of us are recording everything. We're soon going to have these AR/XR glasses. Besides recording audio, they'll be recording visually—your entire ecosystem as you move through the day.

All of this data is being soaked up and made accessible and available, in part, to yourself. But there are going to be companies that are soaking it in, offering to buy it from you to use it to understand what's going on in the world. The world is about to dramatically change in this regard.

Yeah, it goes back to my same point, Peter, about accountability, because you never really asked me if you'd allow me to be recorded or not. Of course, we're recorded on this, but count the number of people whose privacy this one device infringes on.

Imagine a future where that device becomes mandatory if the government decides that this is important for everyone. Think about all of the carbon footprint that a billion of those devices—or 8 billion of them—would mean.

I love technological advancement. I think the question becomes—we should start to call things as they are. I can comfortably say that I grew up in a dictatorship. There's really no doubt about it.

I think we should probably start to think about what we just said: The U.S. is now an experiment. It's not really—I don't think we should continue to call it a democracy. A world where everything is recorded and analyzed is a world with no privacy whatsoever. But I think we lost privacy a long time ago.

Right. I mean, I wonder why we accepted that. Well, I think it's because when you give up privacy, you gain a whole bunch of automagical benefits for yourself, which was the original premise. Now you give up privacy and you get nothing back, perhaps.

Salim, how are you thinking about this? What do you think of my Limitless AI pendant here?

No, by the way, I don't mind. You have my consent, Peter.

Thank you. But I had your consent on this podcast to record you as we are, and forever, for every conversation. I'm just saying the implications of it.

Yeah, no, it's true. But we have to realize we're heading into a world where, as a kid, if you did something silly, the likelihood that it got through to others or was recorded was almost nonexistent. Today we're seeing kids whose college applications are rejected because of some post on Facebook that lives there forever.

There's going to be a future in which everything we're saying and doing is 100% ultimately recorded. We're already there. Big chunks of constitutional rights are falling away as we speak.

In 2015, a study showed that the U.S. is not a functioning democracy in any way, shape, or form. What they meant by that was that there's no amount of public will that can result in legislation. Eighty-four percent of the country believes we should have some form of gun control, and you cannot get gun control passed in any way, shape, or form.

They pointed to a whole bunch of things and found that there's no amount of public will that can result in that. So now we have to think about where we are and where we want to be. It really brings up the big questions, and I think that conversation is not happening enough.

Dave, where you guys—

Yeah, I keep talking about dystopia, but I want to talk about this device, actually. First of all, Jony Ive is an absolute design genius. He's not going to design something dystopian. That's my bet, anyway. I can't wait to see what he comes up with.

This is going to be the always-on device. I think the intelligence of the language models is a total game changer in terms of just a cool, engaging, fun device. If it's done right, it'll help you live a better life and be more aware of your life.

The unexamined life isn't worth living. This is going to be your sounding board. It's not going to have a screen, which I think is great, because your iPhone already has a screen. You can Bluetooth over to the device and look at your iPhone screen if you want a screen.

You can talk to your device through your phone if you want, or you can talk directly to it, but that'll keep the cost down.

So, it should be cheap enough that pretty much everyone on the planet can get one. It will probably be the most impactful device that you buy in your lifetime. The iPhone, or currently the Android phone, would be the reigning life-changing device, but I think this will likely bypass that.

There are so many things that Jony could design here. I just can't wait to see what he comes up with. We know it will be always-on, and we know it will be agent-first. It's going to act like a person. You're going to talk to it like a person, and you're going to feel like it's more like a cuddly teddy bear that you had when you were a kid and less like electronic equipment—your guardian angel there to support and protect you if you need it.

I also think that, strategically, if you look at Fitbit and other past device innovations, you roll them out, try to get market share, and then Apple or Google grabs them and adds them to the operating system of Android and iOS. Then you get crushed. So you have to actually get to market and get a footprint very, very quickly before the big guys come and copy it or try to roll it in with the operating system.

I really think that go-to-market strategy is critical. That's why we want to add 100 million devices in the first iteration, and then we want to add $1 trillion of market cap so that we're a permanent player in the device wars. That's a really good strategy. I'm so excited about that, too.

I'm going to jump into our next topic: chip wars. There's a lot going on in this. So, you mentioned this earlier, Dave. Nvidia projects $1 trillion of annual AI infrastructure spend by 2030. We remind everybody that this year, in 2025, the estimate will be $1 billion a day, which sounds extraordinarily impressive—a billion dollars on the order of $300 billion a year. Let's listen to this quick video from Jensen.

We're going to need a lot more computing, and we're fairly sure now that the world's computing capex is on its way to $1 trillion annually by the end of the decade.

Let's leave it there. That's a lot of capital. You made a point earlier about this being wartime spending, and we're effectively in a private pseudo-war. Can we win the race to AGI, ASI, whatever it might be? Dave, take us from here.

Just to be clear, this is the equivalent amount of dollars, inflation-adjusted, that we spent between 1941 and 1945 during World War II. So it's massive in scale—a huge mobilization. At the time, it was 40% of GDP. Today, it's more like 3% of GDP, so GDP has grown tremendously since then. It's nothing like World War II in terms of everyone getting on it, but it is still an enormous amount of spend.

That's $1 trillion annually, escalating beyond 2030. It still won't be enough because the use cases are bubbling up so quickly, and they get more intelligent and more useful as you iterate more, which means you need more compute. Compute right now is very, very cheap compared to the value and the impact. Protein folding, for example, is just pennies to solve 200 million proteins. It's very, very cheap, but the demand for that is going to be astronomical.

We can't ramp up the spend fast enough to keep up with the use cases. Jensen is exactly right. If anything, it should be that target or more.

Salim?

Well, at least we're unlocking it and making this type of stuff more available in the U.S. and around the world. I think governments will be forced into doing this just to keep up. If you don't have a strategic plan as a country to have a big AI data-center infrastructure, you're going to be left behind very, very quickly.

Mo, you made a comment earlier about the next Avatar movie costing a few thousand dollars rather than a few billion dollars. We've been waiting for Avatar. What are we up to—Avatar 3 coming out soon? Imagine 15,000 versions of Avatar, starring all our favorite friends in there. We're about to see a creative explosion, but we don't have the chip capability. In fact, one of the articles I saw recently said we're not going to be compute-limited; we're going to be energy-limited at the end of the day.

Correct. We'll probably solve that, too. Remember that we're going to apply a lot of intelligence to the way we design chips in a couple of years' time. It is remarkable in every way. Again, remember my point of view: intelligence is a force with no polarity. Applied for good, you get a utopia, so the more of it, the better. Absolutely, no doubt about that.

It is shocking, though, how quickly we're mobilizing on this. When you really think about it, if you just put in place a typical advancement of how much of that hardware will actually be rendered obsolete a few years later because of the advancements in the hardware that comes after, it is such an unusual dynamic. All of us, I think, lived through the dot-com bubble, and we saw that massive expansion mostly redeployed on the internet. This one is just beyond our experience, and the speed of obsolescence is stunning.

Unbelievable. I want to get your opinion here. A couple of weeks ago, we had the entire U.S. AI elite land in Saudi—in Riyadh—and in the Emirates. Ultimately, that was an effort to try and pair the U.S. and the Middle East in the AI world, rather than the Middle East being paired up with China, which was always in the balance in terms of the capital flow and the commitments of capital.

We saw 18,000 of the Blackwell GB300 chips being committed by Jensen to build there. What was it like in Dubai? What was it like in the Middle East? What was going on in the world there on television? How was it being viewed?

I don't know if many people know that, but the largest global AI infrastructure in the world after America and China is in the UAE, which is a tiny country from a size-of-investment point of view. Between the UAE and Saudi Arabia, there is quite an arms race, if you want, in terms of who will build a bigger infrastructure.

It's almost as if Dubai, and now Saudi Arabia, are benefiting from the fact that if you don't have a lot of legacy, you can build quite fast. I think that's definitely something you see in AI infrastructure in general. I do think that it is a very, very clever move to get the Middle East on the American side.

It is not a secret that in every AI meeting that I go to with any ministry or whatsoever, there's always a Chinese side saying, "At least don't take sides." This is a message that is very clear from the Chinese players. I have to say, though, that the expectation from the people of the Middle East is, "We want to see what the U.S. will offer in return," so that the leaders can continue to invest in that way.

It seems to me—and I don't know if that would be speculation on my side—that what we've seen affecting the U.S. Treasury markets after the trade war started sort of requires an influx of funds that stabilizes the markets and the dollar in a way that could only happen with trillions of dollars. I think $4.5 trillion in total was committed here. That's insane. It's a magnificent move.

Most of it is not really announced in terms of what it is, which is why I suspect it would be to support the Treasury markets somehow, or some kind of investment of that sort in the financial markets. The thing, on the other hand, is that this generation of leaders here in the Middle East—Mohammed bin Salman and Mohamed bin Zayed—is the younger generation. They are not as easy to sway to one side or the other because they have grown with enough recognition of their power that they would require a return on that investment.

So let's see how the next move on the chessboard will look.

Speaking of the next move, here's a story from Reuters saying that Chinese tech companies are preparing for an AI future without Nvidia. Alibaba, Tencent, and Baidu are testing Chinese semiconductors to replace Nvidia chips, including those coming out of Huawei.

I just want to address this policy move. If the U.S. starts to restrict exports of technology to China, all this does is cause China to want to innovate around the U.S. We've seen this before, right? We saw this in the telecom industry and in the mobile-phone industry. When we stopped exporting the technology to China, we saw Huawei in particular come in with massive telecom and mobile innovations and steal market share from the U.S.

All of a sudden, the U.S., which should be the dominant provider of this technology to the world, now splits the world with another vendor.

Mo, I'm just going to come back to you on this, and then I'd love to hear from Dave and Salim. Again, I have the privilege of being in touch with both sides, and I can guarantee you there is no coming back from this.

Top-level executives in the Chinese tech world, clearly supported by instructions from the Chinese government, are saying, “We're not going to be dependent on the US's ability to control what chips we get.” Within 3 to 5 years, they'll get to the majority of their needs, but they said the very, very high-end H100 chips are 10 years away.

It is quite staggering when you really think about it because I don't remember the exact number, but they said something like their imports of microchips—including all of the little things, from a child's toy all the way to phones and data centers—exceed their imports of iron and oil combined. That's a massive dollar value.

That basically means that they see massive growth in their economy if they can make those chips locally and replace what they're getting externally from the rest of the world. That once again impacts the Taiwan story and the chip market globally because you're now having a new player that will do things the China way.

Instead of a microchip being X number of dollars, it will now be X number of cents. I have to say, when I saw this conversation for the first time, I thought that was probably one of the dumbest moves by America—to corner them into a place where they are forced to play to their strengths.

We've seen this over and over again with the satellite industry and the launch industry. All of these industries begin to advance because this protectionist move just stimulates the entrepreneurial engine in China to replicate, duplicate, or advance the whole field.

It's interesting that Mo said there's no coming back because that kind of answers my question. What you don't do is poke them and then do nothing. You either win or you don't win. If you're going to embargo—if you're going to basically declare economic war—you'd better declare it to win.

In which case, you have to embargo the chips, but you also have to stop the software flow, the EUV machines, and a few other things. Otherwise, what did you just achieve? All you did is annoy them. If you're going to play, you might as well play to win.

I do think there's a real risk to the US in that we'll say, “They can't make 2-nanometer chips. They can't make 1-nanometer chips.” But it's actually volume that's going to win. If you can manufacture an enormous number of 5-nanometer, or even 10- or 20-nanometer, chips—100 or 1,000 times more of them—that actually works fine for AI and works really, really well, especially for inference-time AI.

There's a danger in assuming that this will work the way it worked with fighter jets. The advanced fighter jet that was slightly better was just unstoppable. This isn't going to be like that. You could win by sheer volume.

When I look at the way the US innovation market works, the reason everybody was in Saudi last week is because that's where the capital is. But don't we have much more capital here in the United States? Jensen was talking about $1 trillion a year of investment, and the US venture capital industry as a whole is one-fifth of that.

Our entire venture capital universe is nowhere near as big as that $1 trillion-a-year investment Jensen was talking about. Where's all our money? It's in pension funds, endowments, and institutions. When you go and talk to them and say, “Hey, why don't you unleash $1 billion?” they say, “No, we don't have an allocation for that. That's above our quota,” or whatever.

Then you go to China or the Middle East, where a much smaller group of decision-makers controls the capital. As Mo was saying, these are great investments. Why is Europe not making these great investments? That's insane.

Europe is destroying itself with its policies today. Literally, that indecision—the inability to make a decision at scale—is absolutely killer in this kind of fast-moving environment. That's why everybody's in Saudi and the UAE: you actually have action in motion.

Mo was right. These investments are absolute no-brainers. They're going to pay off in spades. You're seeing that with the CoreWeave IPO, and you're seeing that with GlobalFoundries. Why did we buy GlobalFoundries from AMD? The chips are going to be in incredible demand, and now we have a foundry.

Salim, do you want to close us out on this one?

Two thoughts. One is that the chip restrictions on China were, I agree with Mo, a really, really dumb idea because it just forces the conversation, and now you've gone down a road you can't come back from.

I note, just as an observation, that 95% of the agricultural drones in the US are Chinese. There's already a huge amount of dependency—forget rare earths—in the engineering and build capability over there in a bunch of sectors. We're playing with fire here.

My bigger hope is that this entire US-China conversation fades away with abundant energy. When you have abundant energy, which is coming very shortly, you can produce lots of things locally at low cost, and you don't need to have this competitive, winner-take-all approach to things.

That's my hope. I may still be living in dreamland, but I'm hoping that's where we get to. This is still the fear-and-scarcity operating software of the human brain from 300 running wild on all of this stuff.

I'm going to continue on this chip conversation. TSMC is accelerating efforts for 1-nanometer production plans and setting up its gigafabs in Taiwan. One nanometer—that's extraordinary.

For reference, the limit of physics is about the diameter of a silicon atom, which is about half a nanometer. We're living in this extraordinary science-fiction universe where we're literally operating at an atomic scale.

I found a few data points. In 2014, we were at 14-nanometer chips from Intel. In 2016, we were at 10 nanometers from Samsung. In 2018, TSMC took the reins at 7 nanometers. They were at 5 nanometers in 2020 and 3 nanometers in 2022. Today, we're at 2 nanometers, and the projection is 1 nanometer by 2030.

One last time, we all started on an 8088. Remember?

I remember the 6502 microprocessor. I was coding in hexadecimal on the 6502.

Yes. I think I did the math. This is 60 trillion times faster than what we can do today in the blink of an eye. Unbelievable—in my lifetime.

And we actually coded some interesting stuff on those machines, right?

Yes, we did. At MIT, I remember the geek kits, Dave. We'd have these giant boxes with gates and chips—AND, NAND, OR, and NOR gates—and we'd literally wire them together.

Actually, Peter, with my geek kit I built an inference-time neural-net accelerator. It had a multiplier in the middle of it, and I didn't appreciate how many wires you had to strip and plug in. I did two—the only time in my life I pulled two back-to-back all-nighters.

You know what we all sound like? We all sound like a bunch of grumpy old men talking about the glory days.

But I love those days. I love it. It was so much fun.

One nanometer, pushing against the limit of physics. Incredible.

Two silicon atoms, or 10 hydrogen atoms. Everything's moving to angstrom terminology now, which is a tenth of a nanometer and the diameter of a hydrogen atom.

One nanometer is the gate width, and that's the physical limit. You can go down to 0.8, maybe, but that's basically the physical limit. The terminology is a little messed up because when they say “1 nanometer,” they're saying it's effectively as if you had 1-nanometer transistors, but they're actually building vertically with the FinFETs.

The gate width is 1 nanometer. It's effectively the same as if you had 1-nanometer transistors, but you're going vertically. That's the end of the line. Now the future belongs to vertical stacking.

Ray Kurzweil was right: we always find a way to continue innovating. That's not going to stop, but it will be in different dimensions. I remember talking to Ralph Merkle last year about this, and he said that as we hit the limits here, we'll go to thermodynamically reversible computation, where we don't generate any heat.

He foresaw a future of using chemical bonds to store the 1s and 0s. That's a whole other level. He figured that would give us 10 orders of magnitude on Moore's law right there—10 billionfold. It's incredible.

I don't think one of the takeaways, though, is that we necessarily need it in order to continue making progress. Some of the more esoteric ideas—for the longest time, gallium arsenide was going to come online and be the next big thing—but it turned out that we just worked around it.

And then carbon nanotubes were going to do whatever—well, that hasn't materialized. So I think what's going to happen here is these will go vertical and massively in scale. We'll get the production costs way down. We'll build enormous data centers horizontally, and we'll also build the chips vertically. That's going to drive innovation for many years to come. Then the next thing may or may not be quantum, and we'll know in a year or 2 whether quantum is going to be the next thing.

I'm going to speed through a few different topics here just to get us through some of the interesting things. We're starting to see AI being used to generate peer-reviewed scientific papers and breakthroughs. We're seeing DeepMind helping us through AlphaEvolve literally solve math records. Dave, I'm hoping that we'll get Alex Wissner-Gross to join us to talk about how AI is going to be solving math, physics, and biology. I think one of the things that's underappreciated is how, over the next 3 years, AI is going to help us accelerate breakthroughs in science beyond anything else we've ever seen before.

Here's another one. This is a demonstration of end-to-end scientific discovery with Robin, a multi-agent system. What we're seeing here is closed-loop scientific robotic and AI systems where an AI proposes an experiment. The robots then run the experiment 24/7, basically in a dark lab, gather the data, feed it back to the AI, which updates its theory and runs the next experiment. We're seeing this in biology for sure. We'll see it in chemistry and materials science. This is another hyper-acceleration in our scientific realm. Thoughts on this, gentlemen? I know that not everybody considers themselves an entrepreneur. What did you say, 16% of America does today?

Adults.

But this is like a field day because all of these areas—I don't want to get into the details of them—are domain-specific. If you can take the current AI, tune it, train it, get proprietary data, and take it down any of these paths, you get miles and miles ahead of generic AI. It's just an entrepreneur's field day these next couple of years.

These are just good case studies. I won't dwell on the specifics; you can read about them later. I'm really excited. This is probably, for me, the biggest small-kid-in-a-wonderland moment, where we can use these AIs to solve really deep physics problems, mathematics problems, and scientific discoveries. A human being trolling through data looking for patterns is terrible. We're bad at that. This is where an AI is really, really good, especially going retroactively and finding all the stuff in experiments that we didn't see in the past. I think I'm unbelievably excited about this.

This is helping humanity across the board. I like to say over and over again—I had this conversation when I was in Hong Kong—that despite the polarity in AI, a breakthrough in biology and longevity in Boston plays equally well in Beijing. It helps us all when humanity is healthier and living longer, more vibrant lives. Mo, anything on the science?

This is my favorite thing ever, AI or not. The possibilities that we have here, just as we go through multidisciplinary sciences, which no human mind has the ability to grasp fully—that is the nature of AI. I think, or at least I dream, that 2026 will be blasted with all of those new discoveries in science, now that we're solving mathematics as well.

The current news media is just playing with your amygdala, delivering negative news over and over and over again every hour into your living room in full color, and you're not hearing all the extraordinary breakthroughs coming our way.

I'm going to move to this last scientific subject, which is one of my favorites. Some of the work being done by Demis Hassabis and others asks: Can we build a full-up virtual AI model of a human cell? Even more importantly, Mo, can we grab a skin cell from you, sequence your DNA, and build a virtual model of Mo?

Why would you ever do that of Dave?

And of Dave, and Salim is a better one. But yes, of each of us. Once you're able to do that, right, we can sequence your genome, put it into a virtual model, and then understand your biology: which medicine, which supplement, and which chemical does or does not work for you, and how exactly it works in your cells. This is the unlock for solving human disease and the limits of longevity. For me, I'm super excited about this, and I think it will happen. It's just a question of time, really.

Yeah, I think it'll happen relatively quickly with AI assist. It is incredibly compute-intensive, so it's a good case study in why those Middle East investments are the biggest no-brainer ever. If you just work backward, the implied amount of computation—and then the benefit of solving virtually every disease—is so overwhelmingly valuable. It unlocks so much capital that you're not wasting on old-age homes or dealing with Alzheimer's or Parkinson's, and it allows humanity to be more productive.

There was a study out of Oxford, London Business School, and Harvard that said for every additional year of health that you give a population, it's worth $38 trillion to the global economy. If you want to solve the U.S. economic issues and China's economic issues, make the population healthier and live longer.

All right, so that wraps up AI and science. Let's talk about one last subject here today, which is something I think about a lot as the father of 2 now-14-year-old boys: reforming education.

I applied to probably around 18 schools, and I was rejected from maybe around 15 of those.

In fact, we have a map. So, in all the places that rejected you—not to make you feel worse looking at this list—well, let's just share your statistics, because I know that factors into college admissions, right? Your GPA and SAT score.

GPA was 4.42 weighted. SAT score was 1590.

Okay. And, in fact, I think we have that on the graphic just so folks can see that.

So the point of this story is: Are we going to move back to meritocracy, where selection and admission into schools are based upon performance and not anything else? It's been a sticky subject coming out of the last 4-plus years on DEI.

Dave, you're deeply embedded in AI and technology at MIT. How do you think about this?

Well, schools are between a rock and a hard place because they clearly had quotas, and now the rules say you're not supposed to do that. They're totally stuck between a rock and a hard place on this topic.

But I don't think they do a great job of choosing whom to let in toward any particular outcome that they're targeting, anyway. There's more to life than a 1590 SAT and a 4.42 GPA. I see a lot of students who are underperforming in the classes I teach just because they need to be creative, build businesses, recruit, and motivate people, and none of that gets measured well by these particular metrics.

I see too many people who are curated to be perfect applicants from age 6, and it doesn't go particularly well. I don't know. I think schools should be allowed to choose with pretty broad brushes what they're trying to achieve with their student body. Hopefully, this doesn't go too far.

Well, I think we have to reinvent the whole university system in the first place. One of the questions I've always had for Salim—you and I both have boys the same age, within a month of each other—is whether university is going to be a thing by the time they get to college age, and what its purpose will be. So, Salim, how are you thinking about this?

I'm in the same mode as with driverless cars. I'm desperately hoping the university system implodes in the next 5 years, before my son has to go, purely because taking a 4-year degree to get credentialed in some domain that you're then supposed to be an active worker in for 40 years before you retire is completely out of date.

The model of a university has not changed in 450 years. It's desperately broken. Deep research is fundamentally incredibly important, so I think that's really killer. I note that the most interesting statistic for me is that more than half the CEOs in Silicon Valley have a liberal arts degree, and I find that really interesting because the different models of how you think drive creativity in product design, et cetera. So there's a vector there to be explored in terms of how to think about all this.

And Mo, my experience is that the majority of leaders and influential individuals out of the Middle East are all coming to the U.S. for their degrees.

No, I think the reality is that I meet more MIT and Stanford graduates here than I do in America most of the time. It is quite staggering, actually.

That’s fantastic. I think it’s definitely revived the top management of the region. Very interestingly, I wonder, though, if Salim’s wish will come true.

I don’t know if the university systems would implode, but I definitely think our belief in universities would. It really is quite an interesting thing, because 4 years in a world that’s moving at 1x speed is very different from a world that’s moving at 10x speed, and that’s what we’re seeing now.

So, if everyone’s going into entrepreneurship, everyone can use Lovable or Claude or whatever to write code or start businesses, and agents are everywhere, you’ll probably see the age of entrepreneurship go to 16 and 14. You may see a very different world. Dave, do you want to talk about that? You’ve seen the shift in terms of the companies that are becoming unicorns a decade earlier.

Yeah, no doubt. But I think universities are about friendships, and the friendships turn into company formations. A lot of universities don’t recognize the degree to which that’s the dominant factor keeping them alive.

If they want to survive this transition, they’ve got to embrace that as what they’re delivering. It’s credentials, and it’s relationships between human beings that are the dominant deliverables to the students. So, they need to be turned into entrepreneur boot camps.

Yeah. Well, here we see another article: “UAE to Make ChatGPT Plus Free to All of Its Citizens.” Again, these are forward-looking moves to help. AI is mandatory education now for ages 6 and up, I think. Crazy, right?

But the statistic here is really important. The stat I heard was that a student with AI is learning subjects 2 to 4 times faster than by going to school, and that’s just going to overwhelm the existing system very quickly. I think this is an awesome move.

Also, consider the concept of a curriculum. We only have so many teachers, so we can only afford 12 or 15 subjects. With AI assistance, you can afford 20,000 to a million different subjects.

Not only are the students self-directing at their own pace, but they’re also learning whatever they think is most relevant to their path, which is so much more effective than the old way. It’s hyperpersonalized education: you’re learning math focused on your favorite sports star, movie star, or favorite scenarios and stories.

I’ll close out with this provocative article that came out: around 5% of Thiel Fellows have become billionaires, from Vitalik to Austin Russell. The Thiel Fellowship is paying you to drop out of college.

So, what is this saying? That our most productive years are being wasted in our university experience instead of starting companies? Fascinating thought. Dave, I’ll start with you.

Well, first of all, the Thiel Fellowship doesn’t try to teach you anything. It just selects you. So, it shows you the degree to which the schools are not selecting them. They’re getting nowhere near a 5% unicorn rate coming out of the schools.

But Thiel, given the abundance of big data that’s out there, can just be a better selection and application process. It covers topics like: Are you self-motivated? Are you high energy? Can you recruit? Do you think through these AI topics? Those are all baked into that selection process.

It’s very viable for these credentials to replace the university degree as the credential that everybody wants. It’s something the schools should really be aware of, but it’s not super hard to put together an AI-assisted analytic that tries to predict who’s going to succeed as an entrepreneur.

That’s all the Thiel Fellowship tries to do. It gives you a little bit of money and encourages you, but what it’s really giving you is the credential.

Yeah, I agree with Dave on this one. I don’t think this is a negative comment on the university system. I think the Thiel Fellowship selects for people who are such outliers that the university system doesn’t accommodate them anyway. I think there’s a systemic issue on the university side, which we’ve talked about already.

More than half of CEOs in Silicon Valley have a liberal arts degree because the different ways of thinking—how to think—are really important in product strategy, company strategy, and so on.

If you’re doing, for example, a master’s degree in neuroscience today, you’re out of date by the time you finish your degree because computational neuroscience is totally taking over the field. So, undergrad and master’s degrees are mostly irrelevant in terms of learning with AI.

If you’re a student learning with AI, you’re moving and learning 2 to 4 times faster than you would be in school. All sorts of things will happen with this. I mentioned that maybe, hopefully, the university system implodes in the next 5 years, before I have to pay for my kid to go to school when he’s 18.

Yeah. I do think education and health care are the 2 massive industries and expenditures for people that are going to be completely disintermediated, disrupted, democratized, and demonetized, I hope.

All right, a lot of amazing stuff. Let’s wrap around the horn here. Thoughts on today’s conversations? Dave, can we start with you?

Yeah, Mo, it’s been fantastic getting your perspective. Well, look, we’re building toward an intentional future from here forward. It’s what we decide to do and what we decide to build.

I think today we got a really good, deep understanding of some of the risks and things we need to start planning for. But I do feel like everything is solvable if we work on it. We’re on exponential time now, so we have a very limited window of time to work on it.

That was one of my great takeaways from today’s discussion. Much appreciated.

Yeah. Just like there is a singularity, and there is an upside utopia and a downside dystopia, I think we should equally weigh our views of the optimistic possibilities and the dangers or risks that we have to address.

I have to say I’m extremely, extremely excited about the scientific breakthroughs that we can see from AI in the next couple of years. I think the most important topic, even though we didn’t cover as much of it today but mentioned it, is AlphaEvolve and the whole idea of self-evolving AIs.

In my mind, this probably is the top topic to keep your eyes on in the next 12 months.

Salim, close us out, buddy.

I think we should have a whole episode just on AlphaEvolve. It’s such an important topic technically, but also philosophically.

I go back to the standard basis for my optimism: technology has always been a major driver of progress in the world. It might be the only major driver of progress we’ve ever seen, as Ray Kurzweil mentions a lot. Now we have AI uplifting all of these other technologies and enabling them. That’s the reason for the huge optimism.

Yeah. Again, I’ve said this before: I think we’re holding 2 potential futures for humanity in superposition.

One is an extraordinary future of abundance, the upleveling of 8 billion humans on the planet, becoming a multiplanetary species—the Star Trek universe. The other one is not quite as pleasant. We’ll call it the dystopian future.

Dave, one of the things that you said that I want to echo is that we have the ability to create an intentional future. This future is not happening to us. We have the ability to guide where it goes.

I think, for all the entrepreneurs listening today, it’s the most important thing that we can do. What is the vision that you want to create in the world? You have the tools now—access to capital, access to compute, access to intelligence—to go make that future happen.

I don’t think it’s ours to abdicate to somebody else. I think we need to take action. A lot is happening. I respect you all deeply, love you all, and am so excited to be on this journey. I look forward to seeing you guys in a week or so.

Thanks for having us.

Thank you guys for putting up with me.